nfsclient

README

NFS Client Input Plugin

The NFS Client input plugin collects data from /proc/self/mountstats. By default, only a limited number of general system-level metrics are collected, including basic read/write counts. If fullstat is set, a great deal of additional metrics are collected, detailed below.

NOTE Many of the metrics, even if tagged with a mount point, are really per-server. Thus, if you mount these two shares: nfs01:/vol/foo/bar and nfs01:/vol/foo/baz, there will be two near identical entries in /proc/self/mountstats. This is a limitation of the metrics exposed by the kernel, not the telegraf plugin.

Global configuration options

In addition to the plugin-specific configuration settings, plugins support additional global and plugin configuration settings. These settings are used to modify metrics, tags, and field or create aliases and configure ordering, etc. See the CONFIGURATION.md for more details.

Configuration

# Read per-mount NFS client metrics from /proc/self/mountstats
[[inputs.nfsclient]]
  ## Read more low-level metrics (optional, defaults to false)
  # fullstat = false

  ## List of mounts to explictly include or exclude (optional)
  ## The pattern (Go regexp) is matched against the mount point (not the
  ## device being mounted).  If include_mounts is set, all mounts are ignored
  ## unless present in the list. If a mount is listed in both include_mounts
  ## and exclude_mounts, it is excluded.  Go regexp patterns can be used.
  # include_mounts = []
  # exclude_mounts = []

  ## List of operations to include or exclude from collecting.  This applies
  ## only when fullstat=true.  Symantics are similar to {include,exclude}_mounts:
  ## the default is to collect everything; when include_operations is set, only
  ## those OPs are collected; when exclude_operations is set, all are collected
  ## except those listed.  If include and exclude are set, the OP is excluded.
  ## See /proc/self/mountstats for a list of valid operations; note that
  ## NFSv3 and NFSv4 have different lists.  While it is not possible to
  ## have different include/exclude lists for NFSv3/4, unused elements
  ## in the list should be okay.  It is possible to have different lists
  ## for different mountpoints:  use mulitple [[input.nfsclient]] stanzas,
  ## with their own lists.  See "include_mounts" above, and be careful of
  ## duplicate metrics.
  # include_operations = []
  # exclude_operations = []

Configuration Options

• fullstat bool: Collect per-operation type metrics. Defaults to false.
• include_mounts list(string): gather metrics for only these mounts. Default is to watch all mounts.
• exclude_mounts list(string): gather metrics for all mounts, except those listed in this option. Excludes take precedence over includes.
• include_operations list(string): List of specific NFS operations to track. See /proc/self/mountstats (the "per-op statistics" section) for complete lists of valid options for NFSv3 and NFSV4. The default is to gather all metrics, but this is almost certainly not what you want (there are 22 operations for NFSv3, and well over 50 for NFSv4). A suggested 'minimal' list of operations to collect for basic usage: ['READ','WRITE','ACCESS','GETATTR','READDIR','LOOKUP','LOOKUP']
• exclude_operations list(string): Gather all metrics, except those listed. Excludes take precedence over includes.

N.B. the include_mounts and exclude_mounts arguments are both applied to the local mount location (e.g. /mnt/NFS), not the server export (e.g. nfsserver:/vol/NFS). Go regexp patterns can be used in either.

References

1. nfsiostat
2. net/sunrpc/stats.c - Linux source code
3. What is in /proc/self/mountstats for NFS mounts: an introduction
4. The xprt: data for NFS mounts in /proc/self/mountstats

Metrics

Fields

• nfsstat
  • bytes (integer, bytes) - The total number of bytes exchanged doing this operation. This is bytes sent and received, including overhead and payload. (bytes = OP_bytes_sent + OP_bytes_recv. See nfs_ops below)
  • ops (integer, count) - The number of operations of this type executed.
  • retrans (integer, count) - The number of times an operation had to be retried (retrans = OP_trans - OP_ops. See nfs_ops below)
  • exe (integer, miliseconds) - The number of miliseconds it took to process the operations.
  • rtt (integer, miliseconds) - The total round-trip time for all operations.
  • rtt_per_op (float, miliseconds) - The average round-trip time per operation.

In addition enabling fullstat will make many more metrics available.

Tags

• All measurements have the following tags:

  • mountpoint - The local mountpoint, for instance: "/var/www"
  • serverexport - The full server export, for instance: "nfsserver.example.org:/export"

• Measurements nfsstat and nfs_ops will also include:

  • operation - the NFS operation in question. READ or WRITE for nfsstat, but potentially one of ~20 or ~50, depending on NFS version. A complete list of operations supported is visible in /proc/self/mountstats.

Additional metrics

When fullstat is true, additional measurements are collected. Tags are the same as above.

NFS Operations

Most descriptions come from Reference and nfs_iostat.h. Field order and names are the same as in /proc/self/mountstats and the Kernel source.

Please refer to /proc/self/mountstats for a list of supported NFS operations, as it changes occasionally.

• nfs_bytes

  • fields:
    • normalreadbytes (int, bytes): Bytes read from the server via read()
    • normalwritebytes (int, bytes): Bytes written to the server via write()
    • directreadbytes (int, bytes): Bytes read with O_DIRECT set
    • directwritebytes (int, bytes): Bytes written with O_DIRECT set
    • serverreadbytes (int, bytes): Bytes read via NFS READ (via mmap())
    • serverwritebytes (int, bytes): Bytes written via NFS WRITE (via mmap())
    • readpages (int, count): Number of pages read
    • writepages (int, count): Number of pages written

• nfs_events (Per-event metrics)

  • fields:
    • inoderevalidates (int, count): How many times cached inode attributes have to be re-validated from the server.
    • dentryrevalidates (int, count): How many times cached dentry nodes have to be re-validated.
    • datainvalidates (int, count): How many times an inode had its cached data thrown out.
    • attrinvalidates (int, count): How many times an inode has had cached inode attributes invalidated.
    • vfsopen (int, count): How many times files or directories have been open()'d.
    • vfslookup (int, count): How many name lookups in directories there have been.
    • vfsaccess (int, count): Number of calls to access(). (formerly called "vfspermission")
    • vfsupdatepage (int, count): Count of updates (and potential writes) to pages.
    • vfsreadpage (int, count): Number of pages read.
    • vfsreadpages (int, count): Count of how many times a group of pages was read (possibly via mmap()?).
    • vfswritepage (int, count): Number of pages written.
    • vfswritepages (int, count): Count of how many times a group of pages was written (possibly via mmap()?)
    • vfsgetdents (int, count): Count of directory entry reads with getdents(). These reads can be served from cache and don't necessarily imply actual NFS requests. (formerly called "vfsreaddir")
    • vfssetattr (int, count): How many times we've set attributes on inodes.
    • vfsflush (int, count): Count of times pending writes have been forcibly flushed to the server.
    • vfsfsync (int, count): Count of calls to fsync() on directories and files.
    • vfslock (int, count): Number of times a lock was attempted on a file (regardless of success or not).
    • vfsrelease (int, count): Number of calls to close().
    • congestionwait (int, count): Believe unused by the Linux kernel, but it is part of the NFS spec.
    • setattrtrunc (int, count): How many times files have had their size truncated.
    • extendwrite (int, count): How many times a file has been grown because you're writing beyond the existing end of the file.
    • sillyrenames (int, count): Number of times an in-use file was removed (thus creating a temporary ".nfsXXXXXX" file)
    • shortreads (int, count): Number of times the NFS server returned less data than requested.
    • shortwrites (int, count): Number of times NFS server reports it wrote less data than requested.
    • delay (int, count): Occurances of EJUKEBOX ("Jukebox Delay", probably unused)
    • pnfsreads (int, count): Count of NFS v4.1+ pNFS reads.
    • pnfswrites (int, count): Count of NFS v4.1+ pNFS writes.

• nfs_xprt_tcp

  • fields:
    • bind_count (int, count): Number of_completely new_ mounts to this server (sometimes 0?)
    • connect_count (int, count): How many times the client has connected to the server in question
    • connect_time (int, jiffies): How long the NFS client has spent waiting for its connection(s) to the server to be established.
    • idle_time (int, seconds): How long (in seconds) since the NFS mount saw any RPC traffic.
    • rpcsends (int, count): How many RPC requests this mount has sent to the server.
    • rpcreceives (int, count): How many RPC replies this mount has received from the server.
    • badxids (int, count): Count of XIDs sent by the server that the client doesn't know about.
    • inflightsends (int, count): Number of outstanding requests; always >1. (See reference #4 for comment on this field)
    • backlogutil (int, count): Cumulative backlog count

• nfs_xprt_udp

  • fields:
    • [same as nfs_xprt_tcp, except for connect_count, connect_time, and idle_time]

• nfs_ops

  • fields (In all cases, the operations tag is set to the uppercase name of the NFS operation, e.g. "READ", "FSINFO", etc. See /proc/self/mountstats for a full list):
    • ops (int, count): Total operations of this type.
    • trans (int, count): Total transmissions of this type, including retransmissions: OP_ops - OP_trans = total_retransmissions (lower is better).
    • timeouts (int, count): Number of major timeouts.
    • bytes_sent (int, count): Bytes sent, including headers (should also be close to on-wire size).
    • bytes_recv (int, count): Bytes received, including headers (should be close to on-wire size).
    • queue_time (int, milliseconds): Cumulative time a request waited in the queue before sending this OP type.
    • response_time (int, milliseconds): Cumulative time waiting for a response for this OP type.
    • total_time (int, milliseconds): Cumulative time a request waited in the queue before sending.
    • errors (int, count): Total number operations that complete with tk_status < 0 (usually errors). This is a new field, present in kernel >=5.3, mountstats version 1.1

Example Output

For basic metrics showing server-wise read and write data.

nfsstat,mountpoint=/NFS,operation=READ,serverexport=1.2.3.4:/storage/NFS ops=600i,retrans=1i,bytes=1207i,rtt=606i,exe=607i 1612651512000000000
nfsstat,mountpoint=/NFS,operation=WRITE,serverexport=1.2.3.4:/storage/NFS bytes=1407i,rtt=706i,exe=707i,ops=700i,retrans=1i 1612651512000000000

For fullstat=true metrics, which includes additional measurements for nfs_bytes, nfs_events, and nfs_xprt_tcp (and nfs_xprt_udp if present). Additionally, per-OP metrics are collected, with examples for READ, LOOKUP, and NULL shown. Please refer to /proc/self/mountstats for a list of supported NFS operations, as it changes as it changes periodically.

nfs_bytes,mountpoint=/home,serverexport=nfs01:/vol/home directreadbytes=0i,directwritebytes=0i,normalreadbytes=42648757667i,normalwritebytes=0i,readpages=10404603i,serverreadbytes=42617098139i,serverwritebytes=0i,writepages=0i 1608787697000000000
nfs_events,mountpoint=/home,serverexport=nfs01:/vol/home attrinvalidates=116i,congestionwait=0i,datainvalidates=65i,delay=0i,dentryrevalidates=5911243i,extendwrite=0i,inoderevalidates=200378i,pnfsreads=0i,pnfswrites=0i,setattrtrunc=0i,shortreads=0i,shortwrites=0i,sillyrenames=0i,vfsaccess=7203852i,vfsflush=117405i,vfsfsync=0i,vfsgetdents=3368i,vfslock=0i,vfslookup=740i,vfsopen=157281i,vfsreadpage=16i,vfsreadpages=86874i,vfsrelease=155526i,vfssetattr=0i,vfsupdatepage=0i,vfswritepage=0i,vfswritepages=215514i 1608787697000000000
nfs_xprt_tcp,mountpoint=/home,serverexport=nfs01:/vol/home backlogutil=0i,badxids=0i,bind_count=1i,connect_count=1i,connect_time=0i,idle_time=0i,inflightsends=15659826i,rpcreceives=2173896i,rpcsends=2173896i 1608787697000000000

nfs_ops,mountpoint=/NFS,operation=NULL,serverexport=1.2.3.4:/storage/NFS trans=0i,timeouts=0i,bytes_sent=0i,bytes_recv=0i,queue_time=0i,response_time=0i,total_time=0i,ops=0i 1612651512000000000
nfs_ops,mountpoint=/NFS,operation=READ,serverexport=1.2.3.4:/storage/NFS bytes=1207i,timeouts=602i,total_time=607i,exe=607i,trans=601i,bytes_sent=603i,bytes_recv=604i,queue_time=605i,ops=600i,retrans=1i,rtt=606i,response_time=606i 1612651512000000000
nfs_ops,mountpoint=/NFS,operation=WRITE,serverexport=1.2.3.4:/storage/NFS ops=700i,bytes=1407i,exe=707i,trans=701i,timeouts=702i,response_time=706i,total_time=707i,retrans=1i,rtt=706i,bytes_sent=703i,bytes_recv=704i,queue_time=705i 1612651512000000000

Documentation

Index

• type NFSClient
  • func (n *NFSClient) Gather(acc telegraf.Accumulator) error
  • func (n *NFSClient) Init() error
  • func (*NFSClient) SampleConfig() string

Types

type NFSClient

type NFSClient struct {
	Fullstat          bool            `toml:"fullstat"`
	IncludeMounts     []string        `toml:"include_mounts"`
	ExcludeMounts     []string        `toml:"exclude_mounts"`
	IncludeOperations []string        `toml:"include_operations"`
	ExcludeOperations []string        `toml:"exclude_operations"`
	Log               telegraf.Logger `toml:"-"`
	// contains filtered or unexported fields
}

func (*NFSClient) Gather

func (n *NFSClient) Gather(acc telegraf.Accumulator) error

func (*NFSClient) Init

func (n *NFSClient) Init() error

func (*NFSClient) SampleConfig

func (*NFSClient) SampleConfig() string